Edge preserving image manipulation in computational photography

Abstract

Computational photography is an emerging field which is the convergence of computer vision, image processing, computer graphics and photography. Its goal is to produce more perceptual representation of our visual world through using computational techniques. Among this field edge preserving image manipulation is widely adopted in many aspects such as image filtering, high dynamic range imaging and non-photorealistic rendering, etc. As human observers are sensitive to image edges when perceiving an image, edge preserving image manipulation will bring observers similar perception between the original image and manipulated one. In the meanwhile, such manipulation will make predesigned effect on the original image to solve specific tasks. An additional advantage of edge preserving image manipulation is that it can be executed very fast in practice, which is suitable for both desktop and mobile computing. In the following we present several methods derived from this manipulation to solve computer vision problems. Image decolorization is widely used in single channel image processing applications such as black and white printer. It is a dimensionality reduction problem which maps color value into grayscale at each pixel location of the color image. The key issue in the image decolorization is how to keep human perception of the color image in the grayscale output. From the edge preserving perspective, we present two methods to address this issue. In the first method we study various kinds of decolorization models and propose one which is robust to improve the state-of-the-art performance. For the second method, we propose a novel edge preserving decolorization method which enables real-time performance for high resolution images. Face sketch synthesis is currently popular in digital entertainment. The state-of-the-art methods involve supervised learning framework which contains an aligned training photo-sketch database. The style of sketches drawn by the artist can be transferred to generate similar sketch under the guidance of input photo. We propose one method which can effectively synthesize face sketches in lower computational cost compared with state-of-the-art methods.

Date of Award	3 Oct 2014
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Qingxiong YANG (Supervisor)